一种基于多层次特征的目标继承提取方法

The Method of Inherited Extraction of Target Based on Multi-level Features

图像的有效分割或提取是图像处理与分析的重要内容,本文使用一种基于多层次特征的方法,针对噪声大,干扰多,光照不均,场景多变,目标形状多变这类图像,自原图像开始,得到各层次的特征,继承地利用低高层次特征逐步提取目标,而又可以回溯前面的原图像或低层次特征信息,继承地对所提取目标优化,实际目标与背景分离.实验表明,该算法可以正确高效地提取目标,有较高的鲁奉性和精度,对具有不同目标大小和信噪比的图像也能得到较好的提取效果.

Due to the complexity of background and asymmetrical illumination, the targets of some images are difficult to be effectively segmented by some routine method. In this paper, an algorithm based on features of levels is proposed and designed, which aims at that kind of pictures with big noises, much interference, uneven illumination as well as the changeable scene and targets of forms. It begins from the original picture, getting the features of every level, and extracting target inheritably from the high or low level feature message. Moreover, it also can track back to the original picture or the features of low level, extracted by recursion. So the target is separated from background. The experiment results indicates, that this algorithm can extract the target correctly with high-efficiency and great precision, and can get good extraction result from the image with different size of target and SNR too.